Most to-epub conversion processes leave a certain amount of garbage or inconsistency behind. One of the best I've found (for OpenOffice/LibreOffice) is
writer2xhmtl and I clean up even after that. It's not so much that it has a problem as that it hands over the detritus that exists in your files that you don't even know exists. For example sometimes there are paragraph and character formats left over/ orphaned as you've manipulated the text. They don't show up in print, but they can upset the xhtml export (and, if nothing else) they are unnecessary clutter in the final file. So, after conversion to epub, I go through the css and xhtml files looking for stuff that doesn't belong there.
Specific issues in the css include things like absolute measurements (inches and pixels) when they should be relative (em or %). Also, with my conversion processor, the css styles are come over with name I know, so when I look in the css I can see when there's something in there that I'm not expecting.
I also make sure all the components are carefully named (rather than abusing UUID strings). I think I have certain obsessive traits, and I like even the parts that the reader doesn't see to look neat and tidy. When you open a typical Smashwords epub it's like walking into a house after a wind-storm has been through, and any minor adjustment you might want to make for your own purposes is more difficult because it's all so messy, so many styles and none of them named in any meaningful way.
Much of this probably doesn't matter to the average reader, but it matters to me, I know it's in there.
